3 * Copyright (C) 2011 Novell Inc.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
11 #include <linux/slab.h>
12 #include <linux/namei.h>
13 #include <linux/file.h>
14 #include <linux/xattr.h>
15 #include <linux/rbtree.h>
16 #include <linux/security.h>
17 #include <linux/cred.h>
18 #include "overlayfs.h"
20 struct ovl_cache_entry
{
24 struct list_head l_node
;
31 struct ovl_dir_cache
{
34 struct list_head entries
;
37 struct ovl_readdir_data
{
38 struct dir_context ctx
;
41 struct list_head
*list
;
42 struct list_head middle
;
50 struct ovl_dir_cache
*cache
;
51 struct ovl_cache_entry cursor
;
52 struct file
*realfile
;
53 struct file
*upperfile
;
56 static struct ovl_cache_entry
*ovl_cache_entry_from_node(struct rb_node
*n
)
58 return container_of(n
, struct ovl_cache_entry
, node
);
61 static struct ovl_cache_entry
*ovl_cache_entry_find(struct rb_root
*root
,
62 const char *name
, int len
)
64 struct rb_node
*node
= root
->rb_node
;
68 struct ovl_cache_entry
*p
= ovl_cache_entry_from_node(node
);
70 cmp
= strncmp(name
, p
->name
, len
);
72 node
= p
->node
.rb_right
;
73 else if (cmp
< 0 || len
< p
->len
)
74 node
= p
->node
.rb_left
;
82 static struct ovl_cache_entry
*ovl_cache_entry_new(const char *name
, int len
,
83 u64 ino
, unsigned int d_type
)
85 struct ovl_cache_entry
*p
;
86 size_t size
= offsetof(struct ovl_cache_entry
, name
[len
+ 1]);
88 p
= kmalloc(size
, GFP_KERNEL
);
90 memcpy(p
->name
, name
, len
);
95 p
->is_whiteout
= false;
102 static int ovl_cache_entry_add_rb(struct ovl_readdir_data
*rdd
,
103 const char *name
, int len
, u64 ino
,
106 struct rb_node
**newp
= &rdd
->root
.rb_node
;
107 struct rb_node
*parent
= NULL
;
108 struct ovl_cache_entry
*p
;
112 struct ovl_cache_entry
*tmp
;
115 tmp
= ovl_cache_entry_from_node(*newp
);
116 cmp
= strncmp(name
, tmp
->name
, len
);
118 newp
= &tmp
->node
.rb_right
;
119 else if (cmp
< 0 || len
< tmp
->len
)
120 newp
= &tmp
->node
.rb_left
;
125 p
= ovl_cache_entry_new(name
, len
, ino
, d_type
);
129 list_add_tail(&p
->l_node
, rdd
->list
);
130 rb_link_node(&p
->node
, parent
, newp
);
131 rb_insert_color(&p
->node
, &rdd
->root
);
136 static int ovl_fill_lower(struct ovl_readdir_data
*rdd
,
137 const char *name
, int namelen
,
138 loff_t offset
, u64 ino
, unsigned int d_type
)
140 struct ovl_cache_entry
*p
;
142 p
= ovl_cache_entry_find(&rdd
->root
, name
, namelen
);
144 list_move_tail(&p
->l_node
, &rdd
->middle
);
146 p
= ovl_cache_entry_new(name
, namelen
, ino
, d_type
);
150 list_add_tail(&p
->l_node
, &rdd
->middle
);
156 void ovl_cache_free(struct list_head
*list
)
158 struct ovl_cache_entry
*p
;
159 struct ovl_cache_entry
*n
;
161 list_for_each_entry_safe(p
, n
, list
, l_node
)
164 INIT_LIST_HEAD(list
);
167 static void ovl_cache_put(struct ovl_dir_file
*od
, struct dentry
*dentry
)
169 struct ovl_dir_cache
*cache
= od
->cache
;
171 list_del(&od
->cursor
.l_node
);
172 WARN_ON(cache
->refcount
<= 0);
174 if (!cache
->refcount
) {
175 if (ovl_dir_cache(dentry
) == cache
)
176 ovl_set_dir_cache(dentry
, NULL
);
178 ovl_cache_free(&cache
->entries
);
183 static int ovl_fill_merge(void *buf
, const char *name
, int namelen
,
184 loff_t offset
, u64 ino
, unsigned int d_type
)
186 struct ovl_readdir_data
*rdd
= buf
;
190 return ovl_cache_entry_add_rb(rdd
, name
, namelen
, ino
, d_type
);
192 return ovl_fill_lower(rdd
, name
, namelen
, offset
, ino
, d_type
);
195 static inline int ovl_dir_read(struct path
*realpath
,
196 struct ovl_readdir_data
*rdd
)
198 struct file
*realfile
;
201 realfile
= ovl_path_open(realpath
, O_RDONLY
| O_DIRECTORY
);
202 if (IS_ERR(realfile
))
203 return PTR_ERR(realfile
);
209 err
= iterate_dir(realfile
, &rdd
->ctx
);
212 } while (!err
&& rdd
->count
);
218 static void ovl_dir_reset(struct file
*file
)
220 struct ovl_dir_file
*od
= file
->private_data
;
221 struct ovl_dir_cache
*cache
= od
->cache
;
222 struct dentry
*dentry
= file
->f_path
.dentry
;
223 enum ovl_path_type type
= ovl_path_type(dentry
);
225 if (cache
&& ovl_dentry_version_get(dentry
) != cache
->version
) {
226 ovl_cache_put(od
, dentry
);
229 WARN_ON(!od
->is_real
&& type
!= OVL_PATH_MERGE
);
230 if (od
->is_real
&& type
== OVL_PATH_MERGE
)
234 static int ovl_dir_mark_whiteouts(struct dentry
*dir
,
235 struct ovl_readdir_data
*rdd
)
237 struct ovl_cache_entry
*p
;
238 struct dentry
*dentry
;
239 const struct cred
*old_cred
;
240 struct cred
*override_cred
;
242 override_cred
= prepare_creds();
243 if (!override_cred
) {
244 ovl_cache_free(rdd
->list
);
249 * CAP_DAC_OVERRIDE for lookup
251 cap_raise(override_cred
->cap_effective
, CAP_DAC_OVERRIDE
);
252 old_cred
= override_creds(override_cred
);
254 mutex_lock(&dir
->d_inode
->i_mutex
);
255 list_for_each_entry(p
, rdd
->list
, l_node
) {
259 if (p
->type
!= DT_CHR
)
262 dentry
= lookup_one_len(p
->name
, dir
, p
->len
);
266 p
->is_whiteout
= ovl_is_whiteout(dentry
);
269 mutex_unlock(&dir
->d_inode
->i_mutex
);
271 revert_creds(old_cred
);
272 put_cred(override_cred
);
277 static inline int ovl_dir_read_merged(struct path
*upperpath
,
278 struct path
*lowerpath
,
279 struct list_head
*list
)
282 struct ovl_readdir_data rdd
= {
283 .ctx
.actor
= ovl_fill_merge
,
289 if (upperpath
->dentry
) {
290 err
= ovl_dir_read(upperpath
, &rdd
);
294 if (lowerpath
->dentry
) {
295 err
= ovl_dir_mark_whiteouts(upperpath
->dentry
, &rdd
);
300 if (lowerpath
->dentry
) {
302 * Insert lowerpath entries before upperpath ones, this allows
303 * offsets to be reasonably constant
305 list_add(&rdd
.middle
, rdd
.list
);
307 err
= ovl_dir_read(lowerpath
, &rdd
);
308 list_del(&rdd
.middle
);
314 static void ovl_seek_cursor(struct ovl_dir_file
*od
, loff_t pos
)
316 struct ovl_cache_entry
*p
;
319 list_for_each_entry(p
, &od
->cache
->entries
, l_node
) {
326 list_move_tail(&od
->cursor
.l_node
, &p
->l_node
);
329 static struct ovl_dir_cache
*ovl_cache_get(struct dentry
*dentry
)
332 struct path lowerpath
;
333 struct path upperpath
;
334 struct ovl_dir_cache
*cache
;
336 cache
= ovl_dir_cache(dentry
);
337 if (cache
&& ovl_dentry_version_get(dentry
) == cache
->version
) {
341 ovl_set_dir_cache(dentry
, NULL
);
343 cache
= kzalloc(sizeof(struct ovl_dir_cache
), GFP_KERNEL
);
345 return ERR_PTR(-ENOMEM
);
348 INIT_LIST_HEAD(&cache
->entries
);
350 ovl_path_lower(dentry
, &lowerpath
);
351 ovl_path_upper(dentry
, &upperpath
);
353 res
= ovl_dir_read_merged(&upperpath
, &lowerpath
, &cache
->entries
);
355 ovl_cache_free(&cache
->entries
);
360 cache
->version
= ovl_dentry_version_get(dentry
);
361 ovl_set_dir_cache(dentry
, cache
);
366 static int ovl_iterate(struct file
*file
, struct dir_context
*ctx
)
368 struct ovl_dir_file
*od
= file
->private_data
;
369 struct dentry
*dentry
= file
->f_path
.dentry
;
375 return iterate_dir(od
->realfile
, ctx
);
378 struct ovl_dir_cache
*cache
;
380 cache
= ovl_cache_get(dentry
);
382 return PTR_ERR(cache
);
385 ovl_seek_cursor(od
, ctx
->pos
);
388 while (od
->cursor
.l_node
.next
!= &od
->cache
->entries
) {
389 struct ovl_cache_entry
*p
;
391 p
= list_entry(od
->cursor
.l_node
.next
, struct ovl_cache_entry
, l_node
);
394 if (!p
->is_whiteout
) {
395 if (!dir_emit(ctx
, p
->name
, p
->len
, p
->ino
, p
->type
))
400 list_move(&od
->cursor
.l_node
, &p
->l_node
);
405 static loff_t
ovl_dir_llseek(struct file
*file
, loff_t offset
, int origin
)
408 struct ovl_dir_file
*od
= file
->private_data
;
410 mutex_lock(&file_inode(file
)->i_mutex
);
415 res
= vfs_llseek(od
->realfile
, offset
, origin
);
416 file
->f_pos
= od
->realfile
->f_pos
;
422 offset
+= file
->f_pos
;
432 if (offset
!= file
->f_pos
) {
433 file
->f_pos
= offset
;
435 ovl_seek_cursor(od
, offset
);
440 mutex_unlock(&file_inode(file
)->i_mutex
);
445 static int ovl_dir_fsync(struct file
*file
, loff_t start
, loff_t end
,
448 struct ovl_dir_file
*od
= file
->private_data
;
449 struct dentry
*dentry
= file
->f_path
.dentry
;
450 struct file
*realfile
= od
->realfile
;
453 * Need to check if we started out being a lower dir, but got copied up
455 if (!od
->is_upper
&& ovl_path_type(dentry
) == OVL_PATH_MERGE
) {
456 struct inode
*inode
= file_inode(file
);
458 realfile
=lockless_dereference(od
->upperfile
);
460 struct path upperpath
;
462 ovl_path_upper(dentry
, &upperpath
);
463 realfile
= ovl_path_open(&upperpath
, O_RDONLY
);
464 smp_mb__before_spinlock();
465 mutex_lock(&inode
->i_mutex
);
466 if (!od
->upperfile
) {
467 if (IS_ERR(realfile
)) {
468 mutex_unlock(&inode
->i_mutex
);
469 return PTR_ERR(realfile
);
471 od
->upperfile
= realfile
;
473 /* somebody has beaten us to it */
474 if (!IS_ERR(realfile
))
476 realfile
= od
->upperfile
;
478 mutex_unlock(&inode
->i_mutex
);
482 return vfs_fsync_range(realfile
, start
, end
, datasync
);
485 static int ovl_dir_release(struct inode
*inode
, struct file
*file
)
487 struct ovl_dir_file
*od
= file
->private_data
;
490 mutex_lock(&inode
->i_mutex
);
491 ovl_cache_put(od
, file
->f_path
.dentry
);
492 mutex_unlock(&inode
->i_mutex
);
502 static int ovl_dir_open(struct inode
*inode
, struct file
*file
)
504 struct path realpath
;
505 struct file
*realfile
;
506 struct ovl_dir_file
*od
;
507 enum ovl_path_type type
;
509 od
= kzalloc(sizeof(struct ovl_dir_file
), GFP_KERNEL
);
513 type
= ovl_path_real(file
->f_path
.dentry
, &realpath
);
514 realfile
= ovl_path_open(&realpath
, file
->f_flags
);
515 if (IS_ERR(realfile
)) {
517 return PTR_ERR(realfile
);
519 INIT_LIST_HEAD(&od
->cursor
.l_node
);
520 od
->realfile
= realfile
;
521 od
->is_real
= (type
!= OVL_PATH_MERGE
);
522 od
->is_upper
= (type
!= OVL_PATH_LOWER
);
523 od
->cursor
.is_cursor
= true;
524 file
->private_data
= od
;
529 const struct file_operations ovl_dir_operations
= {
530 .read
= generic_read_dir
,
531 .open
= ovl_dir_open
,
532 .iterate
= ovl_iterate
,
533 .llseek
= ovl_dir_llseek
,
534 .fsync
= ovl_dir_fsync
,
535 .release
= ovl_dir_release
,
538 int ovl_check_empty_dir(struct dentry
*dentry
, struct list_head
*list
)
541 struct path lowerpath
;
542 struct path upperpath
;
543 struct ovl_cache_entry
*p
;
545 ovl_path_upper(dentry
, &upperpath
);
546 ovl_path_lower(dentry
, &lowerpath
);
548 err
= ovl_dir_read_merged(&upperpath
, &lowerpath
, list
);
554 list_for_each_entry(p
, list
, l_node
) {
558 if (p
->name
[0] == '.') {
561 if (p
->len
== 2 && p
->name
[1] == '.')
571 void ovl_cleanup_whiteouts(struct dentry
*upper
, struct list_head
*list
)
573 struct ovl_cache_entry
*p
;
575 mutex_lock_nested(&upper
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
576 list_for_each_entry(p
, list
, l_node
) {
577 struct dentry
*dentry
;
582 dentry
= lookup_one_len(p
->name
, upper
, p
->len
);
583 if (IS_ERR(dentry
)) {
584 pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
585 upper
->d_name
.name
, p
->len
, p
->name
,
586 (int) PTR_ERR(dentry
));
589 ovl_cleanup(upper
->d_inode
, dentry
);
592 mutex_unlock(&upper
->d_inode
->i_mutex
);